Nuno M. Guerreiro (@nunonmg) 's Twitter Profile
Nuno M. Guerreiro

@nunonmg

Research Scientist at @unbabel, PhD Student at @istecnico. From Lisbon, Portugal 🇵🇹.

ID: 1415338317764349959

linkhttp://nunonmg.github.io calendar_today14-07-2021 15:52:13

192 Tweet

376 Followers

312 Following

Slator (@slatornews) 's Twitter Profile Photo

Researchers from industry and academia have released EuroLLM, the first #LLM that supports all #EU languages 🇪🇺 and several additional languages 🌍 and excels 🚀 at machine #translation. slator.com/first-large-la… Pedro Martins Manuel Faysse Andre Martins Ricardo Rei

Marine Carpuat (@marinecarpuat) 's Twitter Profile Photo

🔎🦙🌍Fine-tuning LLMs with small amounts of parallel data is powerful, but how well do translation capabilities transfer to unsupervised language pairs? Aquia Richburg evaluated TOWER models on 132 translation tasks to find out. Check out his poster Conference on Language Modeling (Wed Session 5)

Nuno M. Guerreiro (@nunonmg) 's Twitter Profile Photo

I'm at Philly for Conference on Language Modeling along with a part of the Tower crew 🗼 Reach out if you want to talk about Tower (I'm also on Whova) or the amazing SARDINE lab in Lisbon (we're hiring post-docs)!

Patrick Fernandes (@psanfernandes) 's Twitter Profile Photo

Turns out MBR works really well not only for specific tasks like MT and summarization but also for general-purpose instruction following! Plus you can distil that test-time compute into your model, to get MBR-level performance with cheap greedy decoding🙂

Nuno M. Guerreiro (@nunonmg) 's Twitter Profile Photo

COLM was amazing — one of my favourite conferences so far! Super happy that we got to share our work on Tower 🗼 and EuroLLM 🇪🇺 w/ so many people Tower's next stop is EMNLP, where we'll present our strongest models. As for EuroLLM, stay tuned for a bigger and better model soon!

COLM was amazing —  one of my favourite conferences so far! Super happy that we got to share our work on Tower 🗼 and EuroLLM 🇪🇺 w/ so many people

Tower's next stop is EMNLP, where we'll present our strongest models. As for EuroLLM, stay tuned for a bigger and better model soon!
Danish Pruthi (@danish037) 's Twitter Profile Photo

There are a couple of full-time openings for pre-doctoral research associates in my group. One of them is specifically for candidates interested in ensuring that large language (and vision) models are geo-culturally inclusive. Email me if you'd be interested.

Unbabel (@unbabel) 's Twitter Profile Photo

💥 Today we’re excited to announce the launch of hubs.li/Q02Y2GpL0 - our new standalone AI solution built for businesses looking to scale quickly with cost-effective translations you can trust. 👇 Learn more about Widn and try it for free. hubs.li/Q02Y2G4q0

Nuno M. Guerreiro (@nunonmg) 's Twitter Profile Photo

Super excited to see the impact of our research and development of Tower in your hands! It's amazing to see Tower powering widn.ai 💥 I'd love to know your feedback about the experience using widn.ai --- we'll be constantly improving our models!

Nuno M. Guerreiro (@nunonmg) 's Twitter Profile Photo

The second, even better and bigger model is now out: EuroLLM-9B 🇪🇺 Ranks as the best open EU-made LLM of its size, proving competitive or superior when going up against models like Meta's Llama 3.1, Qwen 2.5, and Google's Gemma-2. Blog post & models: lnkd.in/d9JJvmd7

Benjamin Minixhofer (@bminixhofer) 's Twitter Profile Photo

We created Approximate Likelihood Matching, a principled (and very effective) method for *cross-tokenizer distillation*! With ALM, you can create ensembles of models from different families, convert existing subword-level models to byte-level and a bunch more🧵

We created Approximate Likelihood Matching, a principled (and very effective) method for *cross-tokenizer distillation*!

With ALM, you can create ensembles of models from different families, convert existing subword-level models to byte-level and a bunch more🧵